Gist compression silently deletes dates, and one prompt line fixes it.

The Sleeping Agent (arXiv 2608.11775) found that gist-based compression preserves relational and event structure while discarding temporal expressions, retention collapsing to 3.05%. Rewriting the abstraction prompt to explicitly retain temporal markers lifts it to 62.39% and recovers +0.314 judge accuracy on temporal questions (95% CI 0.254-0.375), while named-entity and event preservation barely move. Measured over 1,935 matched questions across ten LoCoMo conversations. If you summarize conversation history into long-term memory, audit your compression prompt for this. The loss is invisible until someone asks "when."